[paulbalaji]

Problem

Pair[uint32, [][32]byte] fails Check with array type was not declared before use. Two causes:

• type arguments are validated while instantiations are discovered, before any array is declared. typeArgRule bounded a top-level array without a lookup but sent everything else to one, so [32]byte passed while [][32]byte, []*[4]byte, []Key, map[K][][4]byte and func([]*[4]byte) failed
• arrays are declared where source writes them. A generic field V [][3]T on Box[uint16] holds [3]uint16, which source may never write, so it failed even with the first cause fixed

Solution

• typeArgRule descends through pointers and slices and bounds the arrays it reaches, keeping the pointer and slice rejections.
• Check and the emitter declare the arrays a struct field holds through pointer, slice, map and array types before using the field. Programs whose arrays are all written in source already have them declared, so their C doesn't change.

An oversized or zero-length array inside a type argument now reports that rule's message instead of the declaration error. ERROR_CONTRACT changelog entry included.

Tests

• TestGenericArrayTypeArgs*: two programs, 256 inputs each, native Go vs host C: the issue's Pair, a sorted []Pair lookup, pointers, maps, named arrays, nested slices and generics, inferred and function-typed arguments, generic fields with arrays written nowhere else, bounds and nil traps.
• Real SBF on fogo-v0 and solana-v3: 64 inputs per program, all match, traps roll back.
• 15 forms rejected at 79cf25c now compile and their C builds; storage rules still apply inside type arguments.
• Frozen corpus C unchanged. The five examples/* gosvm projects, a fresh starter and the 4 reference ELFs build byte-identical to 79cf25c. make test passes apart from the known TestStdlib* and TestFullSwapFacts host failures.

Linked work

Closes #8. hyperlane-xyz/hyperlane-monorepo#9852 worked around it by wrapping the slice in a struct. Evidence: results/compiler/2026-10-19-generic-array-type-args. Conflicts with the #7 fix on ERROR_CONTRACT and HANDOFF are mechanical.